CN114154069B

CN114154069B - Information pushing method based on big data information feedback and artificial intelligence prediction system

Info

Publication number: CN114154069B
Application number: CN202111488878.0A
Authority: CN
Inventors: 李烁
Original assignee: Hunan Xianggu Information Technology Co ltd
Current assignee: Hunan Xianggu Information Technology Co ltd
Priority date: 2021-12-07
Filing date: 2021-12-07
Publication date: 2022-08-12
Anticipated expiration: 2041-12-07
Also published as: CN114154069A; CN115114531A; CN115114531B

Abstract

The embodiment of the application provides an information pushing method based on big data information feedback and an artificial intelligence prediction system, key subscription feedback content of subscription feedback data is obtained according to analysis of page operation intention characteristics of the subscription feedback data, the key subscription feedback data is obtained, and then after the key subscription feedback data is determined by guiding conversion of value dimensionality, service pushing information corresponding to a target user is generated for a current development service based on the key subscription feedback data, and the pertinence of service pushing is improved.

Description

Information pushing method based on big data information feedback and artificial intelligence prediction system

Technical Field

The application relates to the technical field of artificial intelligence, in particular to an information pushing method based on big data information feedback and an artificial intelligence prediction system.

Background

E-commerce big data mining can be understood as a technology for mining unknown, potential, useful information and knowledge hidden in the mass e-commerce big data from the mass e-commerce big data. Such information is potentially valuable, understandable, and actionable support decisions that may be of interest to the user, may provide benefits to internet service providers, or may provide a breakthrough for scientific research. At present, data and contents are used as the core of the internet, in both traditional industries and novel industries, if the data and the internet are fused successfully in advance, the hidden rules can be found from big data, and the data and the content can preempt the opportunity, become a mark of technical innovation and obtain benefits.

By mining the E-commerce big data, the behavior preference of the user can be identified, and personalized services are provided, so that more convenient services are better provided for the user of the E-commerce website and the decision of an Internet service provider is guided, for example, subscription guidance is performed on the current development service, so that specific content service information of the current development service is provided for different users in a personalized manner. For example, feedback behavior data of related users may be continuously collected to determine a guidance conversion condition in a subscription guidance process, specific service push information corresponding to a target user is sequentially generated, and how to determine high-reliability key subscription feedback data to improve the pertinence of service push is an urgent technical problem to be solved.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, the present application aims to provide an information pushing method and an artificial intelligence prediction system based on big data information feedback.

In a first aspect, the present application provides an information pushing method based on big data information feedback, which is applied to an artificial intelligence prediction system, and the method includes:

obtaining subscription feedback data of a target user aiming at target subscription guide information associated with a current development service, wherein the subscription feedback data comprises page operation data of the target user under a subscription guide page sequence associated with the target subscription guide information;

acquiring page operation intention distribution of the subscription feedback data, and performing guide conversion value analysis on each page operation intention in the page operation intention distribution to obtain a guide conversion value parameter for representing the guide conversion value of the page operation intention;

obtaining a key page operation intention in the page operation intention distribution according to the guide conversion value parameter of the page operation intention;

obtaining key subscription feedback content in the subscription feedback data according to the key page operation intention, wherein the key page operation intention exists in the page operation intents represented by the key subscription feedback content;

and aggregating the key subscription feedback content to obtain key subscription feedback data, and generating service push information corresponding to the target user for the current development service based on the key subscription feedback data.

In a possible implementation manner of the first aspect, before the step of obtaining subscription feedback data of the target user for the target subscription guidance information associated with the currently developed service, the method further includes:

extracting cloud E-commerce behavior big data related to the current development service from a database of a target user on a preset E-commerce platform, wherein the cloud E-commerce behavior big data represent behavior data of a preference vector of a user preference object;

carrying out preference data tracking on the cloud E-commerce behavior big data according to a preference data tracking model to obtain user preference tracking data in the cloud E-commerce behavior big data generated by the preference data tracking model, wherein the user preference tracking data represents a behavior data part related to a user preference object in the cloud E-commerce behavior big data;

carrying out preference probability prediction on the user preference tracking data according to a preference probability prediction model to obtain a preference probability value of a user preference object in the user preference tracking data generated by the preference probability prediction model, wherein the preference data tracking model and the preference probability prediction model form a user preference analysis model, the preference data tracking model and the preference probability prediction model have a logical association relationship, the preference data tracking model is subjected to convergence optimization according to prediction information of the preference probability prediction model, and the preference probability prediction model is subjected to convergence optimization according to the prediction information of the preference data tracking model;

generating a preference thermodynamic diagram of the target user related to the current development service based on the preference probability value of the user preference object, and pushing target subscription guide information associated with the current development service to the target user according to the preference thermodynamic diagram;

wherein the training of the user preference analysis model comprises:

acquiring reference cloud e-commerce behavior big data, reference user preference tracking data in the reference cloud e-commerce behavior big data and reference preference probability values of user preference objects in the reference cloud e-commerce behavior big data, wherein the reference cloud e-commerce behavior big data represents behavior data of preference vectors of the user preference objects, and the reference user preference tracking data represents behavior data parts related to the user preference objects in the reference cloud e-commerce behavior big data;

carrying out preference data tracking on the reference cloud E-commerce behavior big data based on the preference data tracking model to obtain comparison user preference tracking data in the reference cloud E-commerce behavior big data generated by the preference data tracking model;

performing preference probability prediction on the reference user preference tracking data based on the preference probability prediction model to obtain a first comparison preference probability value of a user preference object in the reference user preference tracking data generated by the preference probability prediction model, and performing preference probability prediction on the comparison user preference tracking data based on the preference probability prediction model to obtain a second comparison preference probability value of the user preference object in the comparison user preference tracking data generated by the preference probability prediction model;

maintaining the model weight of the preference probability prediction model, performing convergence optimization on the preference data tracking model according to the preference tracking cost and the preference probability value cost, maintaining the model weight of the preference data tracking model, and performing convergence optimization on the preference probability prediction model according to the preference probability value cost to obtain a user preference analysis model; wherein the preference tracking cost is calculated from the reference user preference tracking data and the comparative user preference tracking data, and the preference probability value cost is calculated from the first comparative preference probability value, the second comparative preference probability value and the reference preference probability value.

For example, in a possible implementation manner of the first aspect, the maintaining the model weight of the preference probability prediction model, and performing convergence optimization on the preference data tracking model according to the preference tracking cost and the preference probability value cost includes:

maintaining model weights for the preference probability prediction model;

calculating a cost value L1 in the preference tracking cost, calculating a cost value L2 and a cost value L3 in the preference probability value cost, and performing convergence optimization on the preference data tracking model based on the weighted cost values of the cost value L1, the cost value L2 and the cost value L3;

wherein the cost value L1 characterizes loss parameter values for the reference and comparative user preference tracking data, the cost value L2 characterizes loss parameter values for the second and reference preference probability values, and the cost value L3 characterizes loss parameter values for the first and second comparative preference probability values.

For example, in a possible implementation manner of the first aspect, the calculating a cost value L1 in the preference tracking cost includes:

calculating the cost value L1 based on a first preset cost value function according to the reference user preference tracking data and the comparative user preference tracking data, the first preset cost value function being calculated based on a sum of a first preset cost value function and a second preset cost value function;

the calculating of the cost value L2 and the cost value L3 in the preference probability value cost comprises:

calculating the cost value L2 based on a second preset cost value function according to the second comparison preference probability value and the reference preference probability value, wherein the second preset cost value function is calculated according to a third preset cost function;

calculating the cost value L3 based on a third preset cost value function according to the first comparison preference probability value and the second comparison preference probability value, wherein the third preset cost value function is a function of loss parameter values used for calculating the first comparison preference probability value and the second comparison preference probability value;

the optimizing convergence of the preferred data tracking model based on the weighted cost values of the cost value L1, the cost value L2, and the cost value L3 comprises:

calculating a target tracking cost value based on the cost values of the cost value L1, the cost value L2, and the weighted cost value of the cost value L3;

and performing back propagation in the preference data tracking model according to the target tracking cost value to perform convergence optimization on the preference data tracking model.

For example, in a possible implementation manner of the first aspect, the performing backward propagation in the preference data tracking model according to the target tracking cost value performs convergence optimization on the preference data tracking model, including:

respectively performing fusion calculation on the cost value L2 and the cost value L3 in the target tracking cost value and a time domain weight optimization coefficient to obtain a fusion target tracking cost value, wherein the value of the time domain weight optimization coefficient is associated with historical optimization information for performing convergence optimization on the preference data tracking model;

and performing back propagation in the preference data tracking model according to the fusion target tracking cost value to perform convergence optimization on the preference data tracking model.

For example, in a possible implementation manner of the first aspect, the maintaining the model weight of the preference data tracking model, and performing convergence optimization on the preference probability prediction model according to the preference probability value cost includes:

maintaining model weights of the preference data tracking model, and calculating a cost value L4, a cost value L2 and a cost value L3 in the preference probability value costs;

convergence optimization of the preference probability prediction model is performed based on weighted cost values of the cost value L4, the cost value L2, and the cost value L3, the cost value L4 represents loss parameter values of the first comparison preference probability value and the reference preference probability value, the cost value L2 represents loss parameter values of the second comparison preference probability value and the reference preference probability value, and the cost value L3 represents loss parameter values of the first comparison preference probability value and the second comparison preference probability value.

For instance, in one possible implementation of the first aspect, the calculating a cost value L4, a cost value L2, and a cost value L3 of the preference probability value costs comprises:

calculating the cost value L4 based on a fourth preset cost function according to the first comparison preference probability value and the reference preference probability value, wherein the fourth preset cost function is calculated according to a third preset cost function;

the convergence optimizing the preference probability prediction model based on the weighted cost values of the cost value L4, the cost value L2, and the cost value L3 comprises:

calculating a probabilistic predictive cost value based on the cost values L4, L2, and L3;

and carrying out backward propagation in the preference probability prediction model according to the probability prediction cost value to carry out convergence optimization on the preference probability prediction model.

In a second aspect, an embodiment of the present application further provides an information push system based on big data information feedback, where the information push system based on big data information feedback includes an artificial intelligence prediction system and a plurality of electronic commerce service systems in communication connection with the artificial intelligence prediction system;

the artificial intelligence prediction system is configured to:

According to the above aspects, the page operation intention distribution of the subscription feedback data can be obtained, the guiding transformation value analysis is carried out on each page operation intention in the page operation intention distribution to obtain the guiding transformation value parameter of the page operation intention, and the key page operation intention in the page operation intention distribution is obtained according to the guiding transformation value parameter of the page operation intention; according to the method, the key subscription feedback content in the subscription feedback data is obtained according to the key page operation intention, the key subscription feedback content is aggregated to obtain the key subscription feedback data, the key subscription feedback content of the subscription feedback data is obtained according to the analysis of the page operation intention characteristics of the subscription feedback data, the key subscription feedback data is obtained, and then the key subscription feedback data is determined by guiding the conversion value dimension, so that the service push information corresponding to the target user is generated for the current development service based on the key subscription feedback data, and the pertinence of service push is improved.

Drawings

Fig. 1 is a schematic flow chart of an information push method based on big data information feedback according to an embodiment of the present application.

Detailed Description

An information push system 10 based on big data information feedback provided by an embodiment of the present application is described below. The big data information feedback-based information push system 10 may include an artificial intelligence prediction system 100 and an e-commerce service system 200 communicatively connected to the artificial intelligence prediction system 100.

In an embodiment based on independent conception, referring to fig. 1, the artificial intelligence prediction system 100 and the e-commerce service system 200 in the big data information feedback-based information push system 10 may be based on an information push method based on big data information feedback, which is described in the following method embodiments, and the detailed description of the method embodiments may be referred to in the following steps of the artificial intelligence prediction system 100 and the e-commerce service system 200.

Step S101, obtaining subscription feedback data of a target user aiming at target subscription guide information associated with the current development service, and obtaining page operation intention distribution of the subscription feedback data.

In this embodiment, the subscription feedback data includes page operation data of the target user in a subscription guidance page sequence associated with the target subscription guidance information.

For example, continuous page operation behavior data of the subscription feedback data may be obtained, and the continuous page operation behavior data includes a page operation intention distribution of the subscription feedback data.

The page operation intention distribution can be obtained by parsing the relevant user subscribing to the feedback data in the process of executing the subscription feedback data, for example, each page operation intention subscribing to the feedback data can be included in the page operation intention distribution, and each page operation intention is in a continuous feedback phase of a time domain feature space of the subscription feedback data.

For example, the page operation intention distribution of the subscription feedback data may be obtained by a fixed policy resolution of the subscription feedback data. For example, a page operation intention is extracted from the subscription feedback data, a key page operation intention is obtained from the page operation intention, key subscription feedback content is determined according to the key page operation intention, and key subscription feedback data is determined according to the key subscription feedback content.

For example, obtaining the page operation intention distribution of the subscription feedback data may further include: acquiring continuous page operation behavior data of subscription feedback data; according to the continuous page operation behavior data, obtaining each page operation intention of the subscription feedback data and a continuous feedback stage of each page operation intention in a time domain feature space of the subscription feedback data; and performing relation binding on each page operation intention and the associated continuous feedback stage to obtain page operation intention distribution for subscribing feedback data.

The page operation intention distribution can be a page operation intention list subscribed to the feedback data, and the page operation intention information in the page operation intention list comprises intention characteristic information of the page operation intention and continuous feedback stages of the page operation intention. Wherein the continuous feedback phase may include a feedback trigger node and a feedback end node.

The continuous page operation behavior data of the subscription feedback data is information of the page operation intention containing the subscription feedback data, and the specific way of covering the page operation intention is not limited. For example, as described below.

(1): and the continuous page operation behavior data of the subscription feedback data is continuous page sharing transmission data of the subscription feedback data, and the page operation intention distribution is obtained according to the continuous page sharing transmission data of the subscription feedback data.

For example, obtaining each page operation intention of the subscription feedback data according to the continuous page operation behavior data, and the continuous feedback phase of each page operation intention in the time domain feature space of the subscription feedback data may include:

analyzing continuous page sharing and transmitting data subscribed with feedback data into sharing and transmitting characteristic vectors, and taking each sharing and transmitting type in the sharing and transmitting characteristic vectors as a page operation intention; and a continuous feedback stage of each page operation intention in the time domain feature space of the subscription feedback data is obtained according to the time domain feature node range of the time domain feature space of the continuous page sharing transmission data of each page operation intention.

For example, after the persistent page sharing transmission data is parsed into the sharing transmission feature vector, each page operation intention may be determined by using the sharing transmission tag attribute in the sharing transmission feature vector, that is, each sharing transmission type in the sharing transmission feature vector may be determined as one page operation intention. For example, after the persistent page sharing transfer data is parsed into the sharing transfer feature vector, the page operation intents in the sharing transfer feature vector are obtained according to the transfer meanings of the sharing transfer types of the sharing transfer feature vector.

For example, for a scheme of acquiring a page operation intention in real time by continuously page-sharing delivery data, a delivery time domain space of the continuously page-sharing delivery data may be associated with a delivery time domain space of the subscription feedback data, so that the range of time domain feature nodes of the page operation intention in the time domain feature space of the continuously page-sharing delivery data is always the same as the range of the page operation intention in the continuous feedback stage of the time domain feature space of the subscription feedback data. Based on this, after determining each page operation intention, a continuous feedback stage of subscribing the page operation intention to the time domain feature space of the feedback data can be obtained by ranging each page operation intention in the time domain feature nodes of the time domain feature space of the continuous page sharing transmission data. And then, when generating the page operation intention distribution, carrying out relationship binding on the page operation intention and the associated continuous feedback stage so as to obtain the page operation intention distribution subscribing the feedback data.

(2): and the continuous page operation behavior data of the subscription feedback data is continuous page attention behavior data of the subscription feedback data, and the page operation intention distribution is obtained according to the continuous page attention behavior data of the subscription feedback data.

performing attention activity analysis on each continuous page attention behavior data in the subscription feedback data; according to the attention activity analysis data of each continuous page attention behavior data of the subscription feedback data, obtaining a page operation attention object in the subscription feedback data and target continuous page attention behavior data related to the page operation attention object in the subscription feedback data, wherein the page operation attention object on one continuous page attention behavior data is a page operation intention; and a continuous feedback stage of obtaining the time domain feature space of each page operation intention in the subscription feedback data according to the time domain node range of the target continuous page attention behavior data associated with each page operation intention in the subscription feedback data.

For example, performing attention activity resolution on each persistent page attention behavior data in the subscription feedback data may include: and acquiring subscription feedback data, analyzing the subscription feedback data into a continuous page attention behavior data sequence, and performing attention activity analysis on each continuous page attention behavior data in the continuous page attention behavior data sequence.

For example, for other embodiments, the key subscription feedback content may be determined as a range of time-domain nodes of the target continuous page attention behavior data in the subscription feedback data associated with the page operation intent.

analyzing data according to the attention activity of each continuous page attention behavior data of the subscription feedback data to obtain a page operation attention object in the subscription feedback data and target continuous page attention behavior data related to the page operation attention object in the subscription feedback data, wherein the page operation attention object on one continuous page attention behavior data is a page operation intention; determining behavior node information of continuous page attention behavior data of each page operation intention, wherein the behavior node information of the continuous page attention behavior data is used for representing a time domain node range of target continuous page attention behavior data associated with the page operation intention in subscription feedback data;

and a continuous feedback stage of representing the page operation intention by using the behavior node information of the continuous page attention behavior data associated with the page operation intention, and performing relational binding on the page operation intention and the behavior node information of the continuous page attention behavior data associated with the page operation intention to obtain page operation intention distribution subscribing the feedback data.

Correspondingly, acquiring the key subscription feedback content in the subscription feedback data according to the key page operation intention may include: behavior node information of continuous page attention behavior data related to the key page operation intention is obtained from the page operation intention distribution; and acquiring the behavior content of the target continuous page attention behavior data indicated by the behavior node information of the continuous page attention behavior data from the subscription feedback data as the key subscription feedback content in the subscription feedback data.

S102, conducting guide conversion value analysis on each page operation intention in the page operation intention distribution to obtain a guide conversion value parameter for representing the guide conversion value of the page operation intention;

for example, the guided conversion value of the page operation intention in the page operation intention distribution can be analyzed from multiple dimensions,

for example, conducting a guided conversion value analysis on each page operation intention in the page operation intention distribution to obtain a guided conversion value parameter for representing a guided conversion value of the page operation intention may include: acquiring a knowledge graph of the appointed guidance intention; acquiring the entity contact strength of each page operation intention and the appointed guide intention knowledge graph in the page operation intention distribution, and acquiring a guide conversion value parameter for expressing the guide conversion value of the page operation intention according to the entity contact strength associated with the page operation intention, wherein the larger the entity contact strength of the page operation intention and the appointed guide intention knowledge graph is, the larger the guide conversion value parameter of the page operation intention is.

The appointed guidance intention knowledge graph can be configured based on actual service requirements, and a user can set knowledge entities and knowledge entity attributes of the appointed guidance intention knowledge graph.

When the entity connection strength of each page operation intention and the specified guiding intention knowledge graph in the page operation intention distribution is obtained, the entity connection strength of the page operation intention characteristics and the specified guiding intention knowledge graph characteristics can be calculated, and the entity connection strength is used as the entity connection strength of the page operation intention and the specified guiding intention knowledge graph.

For example, entity object division can be performed on the page operation intention characteristics and the designated guidance intention knowledge graph characteristics, statistics of matching entity objects are obtained, and relevant parameter values of the page operation intention and the designated guidance intention knowledge graph are determined according to the statistics of the matching entity objects. Alternatively, the page operation intention feature and the specified guidance intention knowledge graph feature can be extracted through the AI model, and the relevant parameter values of the page operation intention feature and the characteristic variable segment of the specified guidance intention knowledge graph feature are calculated as the relevant parameter values of the page operation intention and the specified guidance intention knowledge graph.

For example, conducting a guided conversion value analysis on each page operation intention in the page operation intention distribution to obtain a guided conversion value parameter for representing a guided conversion value of the page operation intention may include:

acquiring continuous feedback subscription items in subscription feedback data; determining the matching degree of the page operation intention and each subscription item in the subscription feedback data in the page operation intention distribution, and taking the matching degree as a guide conversion value parameter for representing the guide conversion value of the page operation intention, wherein the larger the matching degree is, the larger the guide conversion value parameter of the page operation intention is.

The statistics of the continuous feedback subscription items can be one or more, the matching degree of the page operation intention and the subscription items can be determined by the matching of the page operation intention characteristics and the subscription items expressed by the page operation intention. For example, if a certain page operation intention is a page operation intention expressed by the continuous feedback subscription item, the page operation intention has a greater matching degree with the continuous feedback subscription item; if the continuous feedback subscription item is directly mentioned in the page operation intention characteristic of a certain page operation intention, a certain matching degree exists between the page operation intention and the continuous feedback subscription item.

For example, conducting a guided conversion value analysis on each page operation intention in the page operation intention distribution, and obtaining a guided conversion value parameter for representing a guided conversion value of the page operation intention may include:

combining the page operation intents in the page operation intention distribution to obtain page operation intention clusters, and calculating the entity contact strength between two page operation intents in each page operation intention cluster; and obtaining a guide conversion value parameter of each page operation intention according to the entity connection strength between each page operation intention and the rest page operation intentions in the page operation intention distribution.

The embodiment can determine the entity connection strength between the page operation intents according to the page operation intention characteristics.

For example, calculating the entity contact strength between two page operation intents in each page operation intention cluster may include: calculating a relevant parameter value between two page operation intents in each page operation intention; and acquiring the entity contact strength between the page operation intents according to the relevant parameter values between the page operation intents.

For example, calculating the relevant parameter value between two of the respective page operation intents may include: carrying out entity object division on two page operation intents in the page operation intention cluster, and counting statistics of matched entity objects in the two page operation intents in the page operation intention cluster; counting statistics of the intention components of the two-page operation intentions of the page operation intention cluster; and taking the ratio of the statistic of the matching entity object in the page operation intention cluster to the statistic of the intention component as a relevant parameter value between two page operation intents in the page operation intention cluster.

For example, calculating the relevant parameter value between two page operation intents in each page operation intention cluster may include: extracting characteristic variable fragments for two page operation intents in each page operation intention cluster; and calculating the relevant parameter values of the characteristic variable segments of the two page operation intents in each page operation intention cluster.

For example, the Euclidean distance of the feature variable segments can be used for determining the relevant parameter values.

For example, the guiding conversion value of the page operation intention can be analyzed according to the entity connection strength of each page operation intention and the rest page operation intents.

For example, obtaining the guidance conversion value parameter of each page operation intention according to the entity connection strength between each page operation intention and the remaining page operation intentions in the page operation intention distribution may include:

and calculating the average entity contact strength associated with each page operation intention according to the entity contact strength of each page operation intention and the rest page operation intents in the page operation intention distribution, and taking the average entity contact strength of the page operation intents as a guide conversion value parameter of the page operation intents.

calculating a guide conversion weight of each page operation intention on another page operation intention of the page operation intention cluster according to the entity contact strength of each page operation intention and another page operation intention of the page operation intention cluster in which the page operation intention is located; acquiring a basic guidance conversion value parameter of each page operation intention in the page operation intention distribution; and finally obtaining the final guide conversion value parameters of the page operation intents according to the basic guide conversion value parameters and the guide conversion weights of the other page operation intents of the page operation intention cluster in which the page operation intents are positioned.

Step S103, obtaining a key page operation intention in the page operation intention distribution according to the guiding conversion value parameter of the page operation intention;

for example, after the guidance transformation value parameters of the page operation intentions are obtained, the guidance transformation values of the page operation intentions can be sorted in a descending order according to the guidance transformation value parameters to obtain a page operation intention list; and selecting the page operation intention of the top N in the sequence from the page operation intention list to be determined as the key page operation intention.

Step S104, key subscription feedback content in the subscription feedback data is obtained according to the key page operation intention, wherein the key page operation intention exists in the page operation intention represented by the key subscription feedback content;

for example, the key subscription feedback content may only include the key page operation intention, or may include a part of the page operation intention before and after the key page operation intention, which needs to be determined according to the actual situation of the key page operation intention.

For example, acquiring the key subscription feedback content in the subscription feedback data according to the key page operation intention may include: a continuous feedback stage of obtaining the time domain feature space of the key page operation intention in the subscription feedback data from the page operation intention distribution; according to the continuous feedback stage of the key page operation intention, obtaining the time domain feature node range of the key subscription feedback content related to the key page operation intention in the time domain feature space of the subscription feedback data; and acquiring the key subscription feedback content from the subscription feedback data according to the time domain characteristic node range of the key subscription feedback content.

For example, the time domain feature node range of the key subscription feedback content associated with the key page operation intention can be regarded as the continuous feedback stage of the key page operation intention.

After obtaining the key page operation intention, key subscription feedback content associated with the key page operation intention can be further determined according to the integrity of the transmission meanings of the key page operation intention and the remaining page operation intents.

For example, obtaining a time domain feature node range of the key subscription feedback content associated with the key page operation intention in a time domain feature space of the subscription feedback data according to the continuous feedback stage of the key page operation intention may include: determining a derived predicted page operation intention related to the key page operation intention in the page operation intention distribution, wherein the derived predicted page operation intention is used for forming a complete intention with the key page operation intention; a continuous feedback stage for obtaining the operation intention of the key page and the operation intention of the derived predicted page according to the operation intention of the key page in the page operation intention distribution and the continuous feedback stage for deriving the operation intention of the predicted page; and obtaining the time domain feature node range of the key subscription feedback content associated with the key page operation intention in the time domain feature space of the subscription feedback data in the fusion continuous feedback stage.

For example, the derived predicted page operation intention of the key page operation intention may be determined according to the transfer meaning of the key page operation intention and the surrounding page operation intents, for example, the integrity of the transfer meanings of the key page operation intention and the page operation intents before and after the key page operation intention is analyzed, and the page operation intention forming the complete transfer meaning with the key page operation intention is determined as the derived predicted page operation intention of the key page operation intention.

For example, determining the derived predicted page operation intent associated with the key page operation intent in the distribution of page operation intents may include: acquiring an operation time domain distance between a key page operation intention and a previous page operation intention according to the page operation intention distribution, and if the operation time domain distance is greater than a set time domain distance, determining that no derived predicted page operation intention exists before the key page operation intention; otherwise, determining the derived predicted page operation intention as a selected key page operation intention, and returning to execute the operation of obtaining the operation time domain distance between the key page operation intention and a previous page operation intention according to the page operation intention distribution; acquiring an operation time domain distance between a key page operation intention and a next page operation intention according to the page operation intention distribution, and judging that no derived predicted page operation intention exists after the key page operation intention if the operation time domain distance is greater than a set time domain distance; otherwise, determining the derived predicted page operation intention as the selected key page operation intention, and returning to execute the operation of obtaining the operation time domain distance between the key page operation intention and the next page operation intention from the page operation intention distribution.

For example, the temporal distance may refer to an interval time value.

Step S105, aggregating the key subscription feedback content to obtain key subscription feedback data, and generating service push information corresponding to the target user for the current development service based on the key subscription feedback data.

For example, when the key subscription feedback content is combined, the key subscription feedback data may be obtained by combining content nodes in the subscription feedback data according to the key subscription feedback content.

For example, aggregating the key subscription feedback content to obtain key subscription feedback data may include: according to the continuous feedback stage of the key subscription feedback content, obtaining a content node of the key subscription feedback content in subscription feedback data; and fusing the key subscription feedback content based on the content nodes of the key subscription feedback content to obtain key subscription feedback data.

For example, after aggregating the key subscription feedback content to obtain the key subscription feedback data, the method may further include: obtaining delivery feedback data associated with the subscription feedback data, wherein feedback activities of the delivery feedback data deliver feedback activities associated with the subscription feedback data; and fusing the key subscription feedback data and the transmission feedback data to obtain fused transmission feedback data.

Based on the steps, page operation intention distribution for subscribing feedback data can be obtained, guide conversion value analysis is carried out on each page operation intention in the page operation intention distribution to obtain guide conversion value parameters of the page operation intention, and key page operation intents in the page operation intention distribution are obtained according to the guide conversion value parameters of the page operation intention; according to the method, the key subscription feedback content in the subscription feedback data is obtained according to the key page operation intention, the key subscription feedback content is aggregated to obtain the key subscription feedback data, the key subscription feedback content of the subscription feedback data is obtained according to the analysis of the page operation intention characteristics of the subscription feedback data, the key subscription feedback data is obtained, and then the key subscription feedback data is determined by guiding the conversion value dimension, so that the service push information corresponding to the target user is generated for the current development service based on the key subscription feedback data, and the pertinence of service push is improved.

On the basis of the above description, the following describes in detail the determination manner of the target subscription guidance information of the target user for the current development service association.

Step W102: and acquiring reference cloud e-commerce behavior big data, reference user preference tracking data in the reference cloud e-commerce behavior big data and reference preference probability values of user preference objects in the reference cloud e-commerce behavior big data.

The reference cloud E-commerce behavior big data represents behavior data of a preference vector of a user preference object. For example, the reference cloud e-commerce behavior big data can be live attention behavior data of a user preference object obtained according to an e-commerce live broadcast scene. The user preference object may refer to a pointing object in which a relevant user has a preference behavior (such as a browsing behavior, a clicking behavior, a sharing behavior, and the like) in the e-commerce service flow, such as a certain commodity object, a certain page object, or a certain video object, but is not limited thereto.

The reference user preference tracking data represents a behavior data part related to a user preference object in the reference cloud E-commerce behavior big data. For example, each piece of e-commerce behavior data in the cloud-referenced e-commerce behavior big data corresponds to labeling information, and the labeling information includes a user preference object, a user behavior object, and a scene object. The artificial intelligence prediction system can predict the reference user preference tracking data based on the e-commerce behavior data of the user preference object in the corresponding annotation information. The annotation information may be manually annotated by the user or may be generated by the associated AI model prediction.

The reference preference probability value of the user preference object in the reference cloud E-commerce behavior big data refers to the preference probability value of the user preference object in the reference cloud E-commerce behavior big data. The preference probability value can be manually labeled by the user or generated by prediction of the relevant AI model, and can reflect the interest degree of the user preference object for the relevant user.

Step W104: and carrying out preference data tracking on the reference cloud E-commerce behavior big data based on the preference data tracking model to obtain comparative user preference tracking data in the reference cloud E-commerce behavior big data generated by the preference data tracking model.

The preference data tracking model can locate comparison user preference tracking data related to a user preference object in reference cloud E-commerce behavior big data based on a preference vector of the reference cloud E-commerce behavior big data, and set the dominant value of E-commerce behavior data in the comparison user preference tracking data to 100, and set the dominant value of E-commerce behavior data outside the comparison user preference tracking data to 0, so that a behavior data dominant segmentation region is generated, and then the comparison user preference tracking data is located. For example, the preference data tracking model can be made up of PSPNet and ResNet.

Step W106: and performing preference probability prediction on the reference user preference tracking data based on the preference probability prediction model to obtain a first comparison preference probability value of a user preference object in the reference user preference tracking data generated by the preference probability prediction model, and performing preference probability prediction on the comparison user preference tracking data based on the preference probability prediction model to obtain a second comparison preference probability value of the user preference object in the comparison user preference tracking data generated by the preference probability prediction model.

The preference probability prediction model may predict a preference vector of a user preference object in the passed reference user preference tracking data to generate an associated first comparative preference probability value. And predicting a preference vector for the user preference object in the communicated comparative user preference tracking data to generate an associated second comparative preference probability value.

The preference data tracking model and the preference probability prediction model form a user preference analysis model, the preference data tracking model and the preference probability prediction model have a logical association relationship, and the preference data tracking model can be configured in front of the preference probability prediction model, so that the prediction information of the preference data tracking model can be transmitted to the preference probability prediction model.

Step W108: and carrying out convergence optimization on the preference data tracking model and the preference probability prediction model according to the preference tracking cost and the preference probability value cost to obtain a user preference analysis model.

The preference tracking cost is calculated from the reference user preference tracking data and the comparison user preference tracking data, and the preference probability value cost is calculated from the first comparison preference probability value, the second comparison preference probability value and the reference preference probability value. I.e. the preference tracking cost is calculated based on the prediction information of the preference data tracking model and the preference probability value cost is calculated based on the prediction information of the preference probability prediction model. And carrying out convergence optimization on the preference data tracking model and the preference probability prediction model according to the preference tracking cost and the preference probability value cost, carrying out convergence constraint on the preference data tracking model based on the preference probability prediction model, and carrying out convergence constraint on the preference data tracking model based on the preference data tracking model to carry out convergence optimization on the preference probability prediction model. In the convergence optimization process, the prediction information of the preference data tracking model can improve the precision of the prediction information of the preference probability prediction model, and the prediction information of the preference probability prediction model can learn and train the prediction information of the preference data tracking model, so that the precision of the prediction information of the preference data tracking model is improved.

Based on the above steps, convergence optimization is performed on the preference data tracking model and the preference probability prediction model in accordance with a preference tracking cost calculated based on the prediction information of the preference data tracking model and a preference probability value cost calculated based on the prediction information of the preference probability prediction model, whereby convergence constraint is performed on the preference data tracking model based on the preference probability prediction model to perform convergence optimization on the preference data tracking model, and convergence constraint is performed on the preference data tracking model based on the preference data tracking model to perform convergence optimization on the preference probability prediction model. Therefore, in the convergence optimization process, the accuracy of the prediction information of the preference probability prediction model can be improved by the prediction information of the preference data tracking model, and the prediction information of the preference probability prediction model can be learned and trained to improve the accuracy of the prediction information of the preference data tracking model. Based on the converged preference data tracking model and the preference probability prediction model, better preference prediction performance can be obtained.

Another artificial intelligence-based user preference analysis model training method provided in the embodiments of the present application is described below, which includes the following steps.

Step W202: and acquiring reference cloud e-commerce behavior big data, reference user preference tracking data in the reference cloud e-commerce behavior big data and reference preference probability values of user preference objects in the reference cloud e-commerce behavior big data.

The cloud e-commerce behavior big data is referred to represent behavior data of a preference vector of the user preference object, for example, the cloud e-commerce behavior big data can be live attention behavior data of the user preference object obtained according to an e-commerce live broadcast scene. The reference user preference tracking data represents a behavior data part related to the user preference object in the reference cloud E-commerce behavior big data. For example, each piece of e-commerce behavior data in the cloud-end reference e-commerce behavior big data corresponds to labeling information, the labeling information comprises a user preference object, a user behavior object and a scene object, and the artificial intelligence prediction system can predict the reference user preference tracking data based on the labeling information. The annotation information may be manually annotated by the user or may be generated by the associated AI model prediction. The reference preference probability value refers to a preference probability value of a user preference object in the reference cloud e-commerce behavior big data.

Step W204: and expanding and extending the reference cloud E-commerce behavior big data.

The reference cloud E-commerce behavior big data belongs to a reference behavior data sequence, and the reference behavior data sequence is used for carrying out convergence optimization on a user preference analysis model. And the artificial intelligence prediction system normalizes the reference cloud e-commerce behavior big data to obtain normalized cloud e-commerce behavior big data. And then, expanding and extending the normalized cloud E-commerce behavior big data to obtain expanded and extended cloud E-commerce behavior big data, and adding the expanded and extended cloud E-commerce behavior big data to a reference behavior data sequence.

Step W206: and carrying out preference data tracking on the reference cloud E-commerce behavior big data based on the preference data tracking model to obtain comparative user preference tracking data in the reference cloud E-commerce behavior big data generated by the preference data tracking model.

The preference data tracking model can predict comparative user preference tracking data related to the user preference object in the cloud-end E-commerce behavior big data based on the preference vector of the cloud-end E-commerce behavior big data.

Step W208: a preliminary convergence optimization is performed on the preference data tracking model based on a loss parameter value comparing the user preference tracking data to the reference user preference tracking data.

The artificial intelligence prediction system may predict a cost value L1 based on a first preset cost value function based on comparing the user preference tracking data to the reference user preference tracking data. Then back propagation according to the cost value L1 can perform preliminary convergence optimization on the preference data tracking model. The cost value L1 characterizes a loss parameter value of the reference user preference tracking data versus the comparative user preference tracking data. The first preset cost value function is calculated based on an added value of the first preset cost value function and the second preset cost value function.

Step W210: and performing preference probability prediction on the reference user preference tracking data based on the preference probability prediction model to obtain a first comparison preference probability value of a user preference object in the reference user preference tracking data generated by the preference probability prediction model, and performing preference probability prediction on the comparison user preference tracking data based on the preference probability prediction model to obtain a second comparison preference probability value of the user preference object in the comparison user preference tracking data generated by the preference probability prediction model.

In the above, the preference data tracking model and the preference probability prediction model form a user preference analysis model, the preference data tracking model and the preference probability prediction model have a logical association relationship, and the preference data tracking model can be configured before the preference probability prediction model, so that the prediction information of the preference data tracking model can be transmitted to the preference probability prediction model.

Step W212: and performing preliminary convergence optimization on the preference probability prediction model based on at least one of the loss parameter value of the first comparison preference probability value and the reference preference probability value and the loss parameter value of the second comparison preference probability value and the reference preference probability value.

The artificial intelligence prediction system can predict a cost value L2 based on the second comparison preference probability value and the reference preference probability value and based on a second preset cost value function, and performs back propagation to perform preliminary convergence optimization on the preference probability prediction model according to the cost value L2. The cost value L2 characterizes a loss parameter value of the second comparison preference probability value to the reference preference probability value. The second predetermined cost function is calculated according to a third predetermined cost function.

The artificial intelligence prediction system can predict a cost value L4 based on the first comparison preference probability value and the reference preference probability value and based on the preset cost value function, and performs back propagation to perform preliminary convergence optimization on the preference probability prediction model according to the cost value L4. The cost value L4 characterizes a loss parameter value of the first comparison preference probability value to the reference preference probability value.

For example, the artificial intelligence prediction system can also transmit reference cloud e-commerce behavior big data to the preference probability prediction model to obtain a third comparison preference probability value, and based on the third comparison preference probability value and the reference preference probability value, a fifth generation value can be predicted based on the preset generation value function, and the preference probability prediction model is subjected to preliminary convergence optimization by performing back propagation according to the fifth cost value.

As described above, when the preliminary convergence optimization is performed on the preference data tracking model and the preference probability prediction model, the convergence optimization is performed on the preference data tracking model or the preference probability prediction model alone, and the time sequence before and after the convergence optimization is not specifically limited. In addition, the preference data tracking model and the preference probability prediction model do not need to have a logical association relationship in the whole optimization process. Before the preference data tracking model and the preference probability prediction model are subjected to preliminary convergence optimization, the artificial intelligence prediction system can also perform convergence optimization on the preference data tracking model and the preference probability prediction model based on the disclosed convergence optimization data, so that the preference data tracking model and the preference probability prediction model are initialized.

Step W214: and carrying out convergence optimization on the preference data tracking model and the preference probability prediction model according to the preference tracking cost and the preference probability value cost to obtain a user preference analysis model.

The preference tracking cost is calculated based on the prediction information of the preference data tracking model, and the preference probability value cost is calculated based on the prediction information of the preference probability prediction model. The preference tracking cost is calculated according to the reference user preference tracking data and the comparison user preference tracking data, and the preference probability value cost is calculated according to the first comparison preference probability value, the second comparison preference probability value and the reference preference probability value. And carrying out convergence optimization on the preference data tracking model and the preference probability prediction model according to the preference tracking cost and the preference probability value cost, carrying out convergence constraint on the preference data tracking model based on the preference probability prediction model, and carrying out convergence constraint on the preference data tracking model based on the preference data tracking model to carry out convergence optimization on the preference probability prediction model.

For example, the step W214 may include the following steps W2142 to S2146.

In step W2142, the model weight of the preference probability prediction model is maintained, and the preference data tracking model is subjected to convergence optimization according to the preference tracking cost and the preference probability value cost.

An artificial intelligence prediction system maintains model weights for the preference probability prediction model. And calculates a cost value L1 of the preference tracking cost and determines a cost value L2 and a cost value L3 of the preference probability value cost. The preferred data tracking model is then convergence optimized based on the weighted cost values of cost value L1, cost value L2, and cost value L3. Wherein the cost value L1 characterizes loss parameter values for the reference and comparative user preference tracking data, the cost value L2 characterizes loss parameter values for the second and reference preference probability values, and the cost value L3 characterizes loss parameter values for the first and second comparative preference probability values.

For example, the artificial intelligence prediction system determines a cost value L1, a cost value L2, and a cost value L3, and performs convergence optimization on the preference data tracking model as follows.

The artificial intelligence prediction system may predict a cost value L1 based on a first preset cost value function calculated based on an added value of a first preset cost value function and a second preset cost value function according to the reference user preference tracking data and the comparative user preference tracking data. The artificial intelligence prediction system can predict a cost value L2 based on a second preset cost value function according to the second comparison preference probability value and the reference preference probability value, wherein the second preset cost value function is calculated according to a third preset cost function. The artificial intelligence prediction system can predict the cost value L3 based on a third preset cost value function according to the first comparison preference probability value and the second comparison preference probability value, wherein the third preset cost value function is a function of loss parameter values used for calculating the first comparison preference probability value and the second comparison preference probability value. After the cost value L1, the cost value L2 and the cost value L3 are determined, the artificial intelligence prediction system can predict the target tracking cost value based on the weighted cost values of the cost value L1, the cost value L2 and the cost value L3. And then carrying out backward propagation in the preference data tracking model according to the target tracking cost value to carry out convergence optimization on the preference data tracking model, wherein the preference probability prediction model only carries out forward propagation at the moment.

The first preset cost function, the second preset cost function, and the third preset cost function may be selected according to actual requirements, such as a hinge loss function, a cross entropy loss function, an exponential loss function, and the like, but not limited thereto.

In addition, in order to improve the constraint effect in the training process, the artificial intelligence prediction system performs fusion calculation on a cost value L2 and a cost value L3 in the target tracking cost value and the time domain weight optimization coefficient respectively to obtain a fusion target tracking cost value. The value of the time domain weight optimization coefficient is associated with historical optimization information for convergence optimization of the preference data tracking model. And performing back propagation in the preference data tracking model according to the fusion target tracking cost value to perform convergence optimization on the preference data tracking model.

In step W2144, the model weight of the preference data tracking model is maintained, and the preference probability prediction model is subjected to convergence optimization according to the preference probability value cost.

An artificial intelligence prediction system maintains model weights for the preference data tracking model. And determines a cost value L4, a cost value L2, and a cost value L3 among the preference probability value costs. And then carrying out convergence optimization on the preference probability prediction model based on the weighted cost values of the cost value L4, the cost value L2 and the cost value L3. Wherein the cost value L4 characterizes the loss parameter values of the first and reference preference probability values, the cost value L2 characterizes the loss parameter values of the second and reference preference probability values, and the cost value L3 characterizes the loss parameter values of the first and second preference probability values.

For example, the artificial intelligence prediction system determines a cost value L4, a cost value L2, and a cost value L3, and performs convergence optimization on the preference probability prediction model as follows.

The artificial intelligence prediction system calculates a cost value L4 based on a fourth preset cost value function according to the first comparison preference probability value and the reference preference probability value, wherein the fourth preset cost value function is calculated according to a third preset cost function. And the artificial intelligence prediction system calculates a cost value L2 based on a second preset cost value function according to the second comparison preference probability value and the reference preference probability value, wherein the second preset cost value function is calculated according to a third preset cost function. And the artificial intelligence prediction system calculates a cost value L3 based on a third preset cost value function according to the first comparison preference probability value and the second comparison preference probability value, wherein the third preset cost value function is a function of loss parameter values for calculating the first comparison preference probability value and the second comparison preference probability value. For example, the third preset cost function is the L1 preset cost function. After the cost values L4, L2 and L3 are determined, the artificial intelligence prediction system can predict the probability prediction cost values based on the weighted cost values of the cost values L4, L2 and L3. And then carrying out backward propagation in the preference probability prediction model according to the probability prediction cost value to carry out convergence optimization on the preference probability prediction model, wherein the preference data tracking model only carries out forward propagation at the moment.

In step W2146, the above-described step W2142 and step W2144 are cyclically and alternately executed.

For example, the artificial intelligence prediction system may perform convergence optimization on the preference data tracking model and the preference probability prediction model in a cycle alternating manner during the convergence optimization on the preference data tracking model and the preference probability prediction model. For example, after one convergence optimization of the preference data tracking model, one convergence optimization of the preference probability prediction model is performed, the two convergence optimization processes can be defined as one convergence optimization unit, and the artificial intelligence prediction system alternately performs the convergence optimization on the preference data tracking model and the preference probability prediction model according to the total number of the convergence optimization units. Until the current convergence optimization unit number reaches the total convergence optimization unit number.

In a possible design idea, the weight information of one of the preference data tracking model and the preference probability prediction model is maintained, the convergence optimization is performed on the weight information of the other model, and the convergence optimization is performed alternately on the preference data tracking model and the preference probability prediction model, so that the convergence optimization is performed on the weight information of one of the preference data tracking model and the preference probability prediction model according to the prediction information of the preference data tracking model and the preference probability prediction model, and the generation precision of the user preference analysis model is effectively improved.

In a possible design idea, convergence optimization is performed on the preference data tracking model based on the determined cost value L1, the cost value L2 and the cost value L3, the convergence optimization of the preference data tracking model is realized based on various cost values, and the precision of the preference data tracking model can be effectively improved.

In a possible design idea, the target tracking cost value is calculated based on a first preset cost value function, a second preset cost value function and a third preset cost value function, the target tracking cost value can be calculated rapidly, and the accuracy of the preference data tracking model can be improved by performing convergence optimization on the preference data tracking model based on the target tracking cost value.

In a possible design idea, the training constraint effect of the preference probability prediction model on the preference data tracking model is dynamically adjusted in the convergence optimization process of the preference data tracking model based on the time domain weight optimization coefficient. The method can reduce training errors and further improve the precision of the preference data tracking model while ensuring the rapid training and optimization of the preference data tracking model.

In a possible design idea, convergence optimization is performed on the preference probability prediction model based on the determined cost value L4, the cost value L2 and the cost value L3, the convergence optimization of the preference probability prediction model is realized based on various cost values, and the precision of the preference probability prediction model can be effectively improved.

In a possible design idea, the probability prediction cost value is calculated based on a fourth preset cost value function, a second preset cost value function and a third preset cost value function, so that the probability prediction cost value can be rapidly calculated

And the probability prediction cost value is obtained, and the accuracy of the preference probability prediction model can be improved by carrying out convergence optimization on the preference probability prediction model based on the probability prediction cost value.

The information pushing method based on big data information feedback provided by the embodiment of the application is described below, and includes the following steps.

Step W302: and extracting cloud E-commerce behavior big data related to the current development service from a database of the target user on a preset E-commerce platform.

The cloud E-commerce behavior big data represents behavior data to be subjected to preference probability value generation. The cloud E-commerce behavior big data represents behavior data of a preference vector of a user preference object. For example, the cloud e-commerce behavior big data represents live attention behavior data of a user preference object obtained according to an e-commerce live broadcast scene.

Step W304: and carrying out preference data tracking on the cloud E-commerce behavior big data according to a preference data tracking model to obtain user preference tracking data in the cloud E-commerce behavior big data generated by the preference data tracking model.

The preference data tracking model can predict user preference tracking data associated with the user preference object in the cloud E-commerce behavior big data based on the preference vector of the cloud E-commerce behavior big data.

Step W306: and carrying out preference probability prediction on the user preference tracking data according to a preference probability prediction model to obtain a preference probability value of a user preference object in the user preference tracking data generated by the preference probability prediction model.

The preference probability prediction model can predict the preference vector of the user preference object in the transmitted user preference tracking data, and further generate the preference probability value of the user preference object in the associated user preference tracking data. The preference data tracking model and the preference probability prediction model form a user preference analysis model, the preference data tracking model and the preference probability prediction model have a logical association relationship, and the preference data tracking model can be configured in front of the preference probability prediction model. And the preference data tracking model is subjected to convergence optimization according to the prediction information of the preference probability prediction model, and the preference probability prediction model is subjected to convergence optimization according to the prediction information of the preference data tracking model. The user preference analysis model may be an artificial intelligence network model optimized for model convergence based on the aforementioned steps W202 to S214.

And step W308, generating a preference thermodynamic diagram related to the current development service of the target user based on the preference probability value of the user preference object, and pushing target subscription guidance information related to the current development service to the target user according to the preference thermodynamic diagram.

For example, the preference thermodynamic diagram of the target user in relation to the currently developed service may be used to characterize the variation data of the preference probability value of each user preference object in the time dimension.

For example, in the process of pushing the target subscription guidance information associated with the currently developed service to the target user according to the preference thermodynamic diagram, a user preference object with a continuously rising preference probability value in a preset time period may be determined as a target user preference object, then, based on the latest preference probability value of each target user preference object, descending order sorting is performed, and subscription guidance pages corresponding to the target user preference object are respectively pushed to the user terminals corresponding to the target user according to the sorting order.

Based on the steps, the accuracy of the prediction information of the preference probability prediction model is improved by the prediction information of the preference data tracking model, the prediction information of the preference probability prediction model can learn and train the prediction information of the preference data tracking model, the accuracy of the prediction information of the preference data tracking model is improved, better preference prediction performance can be obtained based on the converged preference data tracking model and the preference probability prediction model, preference probability prediction is carried out on the preference tracking data of the user, a preference thermodynamic diagram related to the current development service of the target user is generated based on the preference probability value of a preference object of the user, and target subscription guidance information related to the current development service is pushed to the target user according to the preference thermodynamic diagram, so that the subscription guidance accuracy of the current development service is improved.

In an embodiment that may be based on independent concepts, the artificial intelligence prediction system 100 may include: a processor 101 and a machine-readable storage medium 102. Wherein, the machine-readable storage medium 102 is used for storing a program that supports the artificial intelligence prediction system 100 to execute the information push method based on big data information feedback provided in any one of the foregoing embodiments, and the processor 101 is configured to execute the program stored in the machine-readable storage medium 102.

The program comprises one or more computer instructions, wherein the one or more computer instructions, when executed by the processor 101, enable all or part of the steps of any of the foregoing embodiments.

The architecture of the artificial intelligence prediction system 100 may further include a communication unit 103, which is used for the artificial intelligence prediction system 100 to communicate with other devices or communication networks (e.g., the e-commerce service system 200).

In addition, the present application provides a computer storage medium for storing computer software instructions for the artificial intelligence prediction system 100, which includes a program for executing the information push method based on big data information feedback in any one of the foregoing method embodiments.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims

1. An information pushing method based on big data information feedback is applied to an artificial intelligence prediction system, and the method comprises the following steps:

aggregating the key subscription feedback content to obtain key subscription feedback data, and generating service push information corresponding to the target user for the current development service based on the key subscription feedback data;

the step of analyzing the guide conversion value of each page operation intention in the page operation intention distribution to obtain a guide conversion value parameter for representing the guide conversion value of the page operation intention includes:

acquiring a specified guidance intention knowledge graph, and acquiring entity contact strength of each page operation intention and the specified guidance intention knowledge graph in page operation intention distribution; obtaining a guidance conversion value parameter for representing the guidance conversion value of the page operation intention according to the entity contact strength associated with the page operation intention, wherein the larger the entity contact strength of the page operation intention and the specified guidance intention knowledge graph is, the larger the guidance conversion value parameter of the page operation intention is;

or combining the page operation intents in the page operation intention distribution to obtain page operation intention clusters, and calculating the entity contact strength between two page operation intents in each page operation intention cluster; obtaining a guide conversion value parameter of each page operation intention according to the entity contact strength between each page operation intention and the rest page operation intentions in the page operation intention distribution;

the obtaining of the page operation intention distribution of the subscription feedback data includes:

acquiring continuous page operation behavior data of subscription feedback data;

according to the continuous page operation behavior data, obtaining each page operation intention of the subscription feedback data and a continuous feedback stage of each page operation intention in a time domain feature space of the subscription feedback data;

and performing relation binding on each page operation intention and the associated continuous feedback stage to obtain page operation intention distribution for subscribing feedback data.

2. The method according to claim 1, wherein the continuous page operation behavior data is continuous page sharing delivery data of the subscription feedback data, and the obtaining of each page operation intention of the subscription feedback data according to the continuous page operation behavior data and the continuous feedback phase of each page operation intention in the time domain feature space of the subscription feedback data comprise:

analyzing the continuous page sharing transmission data into sharing transmission characteristic vectors, and taking each sharing transmission type in the sharing transmission characteristic vectors as a page operation intention;

and a continuous feedback stage of each page operation intention in the time domain feature space of the subscription feedback data is obtained according to the time domain feature node range of the time domain feature space of the continuous page sharing transmission data of each page operation intention.

3. The method for pushing information based on big data information feedback according to claim 1, wherein the continuous page operation behavior data is continuous page attention behavior data in the subscription feedback data, and the continuous feedback phase of each page operation intention in the time domain feature space of the subscription feedback data according to the continuous page operation behavior data includes:

performing attention activity analysis on each continuous page attention behavior data in the subscription feedback data;

according to the attention activity analysis data of each continuous page attention behavior data of the subscription feedback data, obtaining a page operation attention object in the subscription feedback data and target continuous page attention behavior data related to the page operation attention object in the subscription feedback data, wherein the page operation attention object on one continuous page attention behavior data is a page operation intention;

and a continuous feedback stage of obtaining the time domain feature space of each page operation intention in the subscription feedback data according to the time domain node range of the target continuous page attention behavior data associated with each page operation intention in the subscription feedback data.

4. The method for pushing information based on big data information feedback according to any one of claims 1 to 3, wherein the obtaining of the key subscription feedback content in the subscription feedback data according to the key page operation intention includes:

a continuous feedback stage of obtaining the key page operation intention in the time domain feature space of the subscription feedback data from the page operation intention distribution;

acquiring an operation time domain distance between a key page operation intention and a previous page operation intention according to the page operation intention distribution, and judging that no derived predicted page operation intention exists before the key page operation intention if the operation time domain distance is greater than a set time domain distance; if the operation time domain distance is not larger than the set time domain distance, determining the derived predicted page operation intention as a selected key page operation intention, and returning to execute the operation of obtaining the operation time domain distance between the key page operation intention and a previous page operation intention according to the page operation intention distribution;

acquiring an operation time domain distance between a key page operation intention and a next page operation intention according to the page operation intention distribution, and judging that no derived predicted page operation intention exists behind the key page operation intention if the operation time domain distance is greater than a set time domain distance; if the operation time domain distance is not larger than the set time domain distance, determining the derived predicted page operation intention as a selected key page operation intention, and returning to execute the operation of obtaining the operation time domain distance between the key page operation intention and a post-positioned page operation intention according to the page operation intention distribution, wherein the derived predicted page operation intention is used for forming a complete intention with the key page operation intention;

obtaining a fusion continuous feedback stage of the key page operation intention and the derived predicted page operation intention according to the continuous feedback stage of the key page operation intention and the derived predicted page operation intention in the page operation intention distribution;

obtaining a time domain feature node range of key subscription feedback content associated with the key page operation intention in a time domain feature space of the subscription feedback data in the fusion continuous feedback stage;

and acquiring the key subscription feedback content from the subscription feedback data according to the time domain feature node range of the key subscription feedback content.

5. The method for pushing information based on big data information feedback according to claim 1, wherein the obtaining of the guidance conversion value parameter of each page operation intention according to the entity contact strength between each page operation intention and the remaining page operation intentions in the page operation intention distribution comprises:

calculating a guide conversion weight of each page operation intention on another page operation intention of the page operation intention cluster according to the entity contact strength of each page operation intention and another page operation intention of the page operation intention cluster;

acquiring basic guidance conversion value parameters of all page operation intents in the page operation intention distribution;

and obtaining the final guide conversion value parameter of each page operation intention according to the base guide conversion value parameter and the guide conversion weight of the other page operation intention of the page operation intention cluster in which each page operation intention is positioned.

6. The big data information feedback-based information pushing method according to claim 1, wherein after aggregating the key subscription feedback content to obtain key subscription feedback data, further comprising:

obtaining delivery feedback data associated with subscription feedback data, wherein the feedback activities of the delivery feedback data deliver feedback activities associated with the subscription feedback data;

fusing the key subscription feedback data and the delivery feedback data to obtain fused delivery feedback data;

or, according to the continuous feedback stage of the key subscription feedback content, obtaining a content node of the key subscription feedback content in the subscription feedback data;

and fusing the key subscription feedback content according to the content node to obtain key subscription feedback data.

7. The method for pushing information based on big data information feedback according to claim 1, wherein before the step of obtaining the subscription feedback data of the target user for the target subscription guidance information associated with the currently developed service, the method further comprises:

wherein the training of the user preference analysis model comprises:

maintaining the model weight of the preference probability prediction model, performing convergence optimization on the preference data tracking model according to preference tracking cost and preference probability value cost, maintaining the model weight of the preference data tracking model, and performing convergence optimization on the preference probability prediction model according to the preference probability value cost to obtain a user preference analysis model; wherein the preference tracking cost is calculated from the reference user preference tracking data and the comparative user preference tracking data, and the preference probability value cost is calculated from the first comparative preference probability value, the second comparative preference probability value and the reference preference probability value.

8. An artificial intelligence prediction system, which is characterized by comprising a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores a computer program, and the computer program is loaded and executed based on the processor to implement the big data information feedback-based information pushing method according to any one of claims 1 to 7.